GANs as Gradient Flows that Converge

被引:0
作者
Huang, Yu-Jui [1 ]
Zhang, Yuchong [2 ]
机构
[1] Univ Colorado, Dept Appl Math, Boulder, CO 80309 USA
[2] Univ Toronto, Dept Stat Sci, Toronto, ON M5G 1Z5, Canada
基金
加拿大自然科学与工程研究理事会; 美国国家科学基金会;
关键词
unsupervised learning; generative adversarial networks; distribution-dependent ODEs; gradient flows; nonlinear Fokker-Planck equations;
D O I
暂无
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
This paper approaches the unsupervised learning problem by gradient descent in the space of probability density functions. A main result shows that along the gradient flow induced by a distribution-dependent ordinary differential equation (ODE), the unknown data distribution emerges as the long-time limit. That is, one can uncover the data distribution by simulating the distribution-dependent ODE. Intriguingly, the simulation of the ODE is shown equivalent to the training of generative adversarial networks (GANs). This equivalence provides a new "cooperative" view of GANs and, more importantly, sheds new light on the divergence of GANs. In particular, it reveals that the GAN algorithm implicitly minimizes the mean squared error (MSE) between two sets of samples, and this MSE fitting alone can cause GANs to diverge. To construct a solution to the distribution-dependent ODE, we first show that the associated nonlinear Fokker-Planck equation has a unique weak solution, by the Crandall-Liggett theorem for differential equations in Banach spaces. Based on this solution to the Fokker-Planck equation, we construct a unique solution to the ODE, using Trevisan's superposition principle. The convergence of the induced gradient flow to the data distribution is obtained by analyzing the Fokker-Planck equation.
引用
收藏
页数:40
相关论文
共 39 条
  • [1] Transport equation and Cauchy problem for BV vector fields
    Ambrosio, L
    [J]. INVENTIONES MATHEMATICAE, 2004, 158 (02) : 227 - 260
  • [2] Ambrosio Luigi, 2005, Lectures in Mathematics ETH Zurich
  • [3] Ansari A. F., 2021, INT C LEARN REPR
  • [4] Ansari Abdul Fatir, 2020, IEEE CVF C COMP VIS
  • [5] Arjovsky M., 2017, stat, P1
  • [6] Arjovsky M, 2017, PR MACH LEARN RES, V70
  • [7] Barbu V, 2010, SPRINGER MONOGR MATH, P1, DOI 10.1007/978-1-4419-5542-5
  • [8] FROM NONLINEAR FOKKER-PLANCK EQUATIONS TO SOLUTIONS OF DISTRIBUTION DEPENDENT SDE
    Barbu, Viorel
    Roeckner, Michael
    [J]. ANNALS OF PROBABILITY, 2020, 48 (04) : 1902 - 1920
  • [9] Bińkowski M, 2021, Arxiv, DOI [arXiv:1801.01401, DOI 10.48550/ARXIV.1801.01401, 10.48550/arXiv:1801.01401]
  • [10] BREZIS H, 1979, J MATH PURE APPL, V58, P153